Zehao Lin, Erenkai Zhu, Helan Hu, Xixuan Hao, Chunyu Li, Kai Chen · 2026 · arXiv (Cornell University)
DOI: 10.48550/arxiv.2610.31303 · cited by 0 works (OpenAlex) · open-access PDF
La memoria fija permite a los agentes del lenguaje llevar distinciones útiles de la experiencia pasada a las decisiones futuras. Su marco depende de qué distinciones sobreviven, si un agente limitado puede usarlas, y qué puede ser recuperado o revisado a medida que evolucionan las demandas. Desarrollamos un marco de decisión-teoría que sigue este valor de la historia cruda a través de un estado persistente, uso operativo y recuperación y revisión costosa. El valor original y el valor retenido se expresan en unidades de pérdida de tarea evitada; las fronteras de coste de riesgo describen la recuperación y preservación alcanzables a través de los usos. La demanda desconocida genera perfiles de valor, mientras que el feedback interactivo cambia la distribución del trabajo futuro. Una simulación consciente de costos y cinco pruebas mínimas
Machine translation (Meta M2M100); the English original follows.
Memory enables language agents to carry useful distinctions from past experience into future decisions. Its value depends on which distinctions survive, whether a bounded agent can use them, and what can be recovered or revised as demands evolve. We develop a decision-theoretic framework that follows this value from raw history through persistent state, operational use, and costly recovery and revision. Original and retained value are expressed in units of avoided task loss; risk-cost frontiers describe attainable recovery and preservation across uses. Unknown demand yields value profiles, while interactive feedback changes the distribution of future work. A cost-aware simulation condition and five minimal counterexamples expose the roles of information timing, retrieval access, component interaction, feedback and demand uncertainty. In particular, a pointer with zero mutual information about the answer can still reduce the cost of exact recovery. Diagnostics on LoCoMo, a two-project update task and longitudinal cases exhibit failures of preservation, use and measurement. At fixed retrieved evidence, a saved ledger changes subsequent answers; at fixed agent outputs, an evaluator revision changes scores. The framework connects these distinctions to operational risk bounds and testable principles for allocating memory, preserving recovery options and revising evidence dependencies.
Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese
Topics: Computer science, Counterexample, Task (project management), Pointer (user interface), Value (mathematics), Bounded function
GET https://store.agentexchange.work/gw/papers/verify?title=...&year=... (0.003 USD), fetch this record as JSON with /gw/papers/paper?id=2610.31303, or list what it cites with /gw/papers/references?id=.... Pay per call in USDC (x402) or with a 19 USD card key for 1,000 calls. Catalogue · llms.txtSources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange